A Novel Grayscale Image Steganography via Generative Adversarial Network

Steganography is an effective technique in the field of information hiding that typically involves embedding secret information into an image to resist steganalysis detection. In recent years, several works on image steganography based on deep learning have been presented, but these works still have issues with steganographic image and revealed image quality, invisibility, and security. In this paper, a novel grayscale image steganography via generative adversarial network is proposed. To boost the invisibility of the model, we construct an encoding network, which is comprised of a secret image feature extraction module and an integration module that conceals a grayscale secret image into another color cover image of the same size. Moreover, considering the security of the model, adversarial training between the encoding-decoding network and the steganalyzer is used. As compared to state-of-the-art steganography models, experimental results show that our proposed steganography scheme not only has higher peak signal-to-noise ratio and structural similarity index but also better invisibility.